Merge pull request #869 from jkottiku/master

Add gpu index in json log
ROCm · Dec 20, 2024 · d9d7201 · d9d7201
2 parents a74ec5a + 71bcfe4
commit d9d7201
Show file tree

Hide file tree

Showing 8 changed files with 53 additions and 12 deletions.
diff --git a/gpup.so/src/action.cpp b/gpup.so/src/action.cpp
@@ -488,9 +488,15 @@ int gpup_action::run(void) {
     if (bjson){
       json_root_node = json_node_create(std::string(module_name),
           action_name.c_str(), rvs::logresults);
-      if(json_root_node)
+      if(json_root_node) {
         rvs::lp::AddString(json_root_node, RVS_JSON_LOG_GPU_ID_KEY, std::to_string(gpu_id[i]));
+
+        uint16_t gpu_index = 0;
+        rvs::gpulist::gpu2gpuindex(gpu_id[i], &gpu_index);
+        rvs::lp::AddString(json_root_node, RVS_JSON_LOG_GPU_IDX_KEY, std::to_string(gpu_index));
+      }
     }
+
     // properties values
     sts = property_get_value(gpu_id[i]);
     sts = property_io_links_get_value(gpu_id[i]);

diff --git a/include/rvs_key_def.h b/include/rvs_key_def.h
@@ -63,5 +63,6 @@
 #define YAML_DEVICE_PROP_DELIMITER      " "
 
 #define RVS_JSON_LOG_GPU_ID_KEY         "gpu_id"
+#define RVS_JSON_LOG_GPU_IDX_KEY        "gpu_index"
 
 #endif  // INCLUDE_RVS_KEY_DEF_H_
diff --git a/include/rvs_util.h b/include/rvs_util.h
@@ -137,6 +137,7 @@ bool fetch_gpu_list(int hip_num_gpu_devices, map<int, uint16_t>& gpus_device_ind
 void getBDF(int idx ,unsigned int& domain,unsigned int& bus,unsigned int& device,unsigned int& function);
 int display_gpu_info(void);
 void *json_list_create(std::string lname, int log_level);
+
 template <typename... KVPairs>
 void log_to_json(action_descriptor desc, int log_level, KVPairs...  key_values ) {
         std::vector<std::string> kvlist{key_values...};
@@ -146,8 +147,13 @@ void log_to_json(action_descriptor desc, int log_level, KVPairs...  key_values )
     void *json_node = json_node_create(desc.module_name,
         desc.action_name.c_str(), log_level);
     if (json_node) {
-      rvs::lp::AddString(json_node, "gpu_id",
-          std::to_string(desc.gpu_id));
+
+      rvs::lp::AddString(json_node, "gpu_id", std::to_string(desc.gpu_id));
+
+      uint16_t gpu_index = 0;
+      rvs::gpulist::gpu2gpuindex(desc.gpu_id, &gpu_index);
+      rvs::lp::AddString(json_node, "gpu_index", std::to_string(gpu_index));
+
       for (int i =0; i< kvlist.size()-1; i +=2){
           rvs::lp::AddString(json_node, kvlist[i], kvlist[i+1]);
       }

diff --git a/mem.so/src/rvs_memworker.cpp b/mem.so/src/rvs_memworker.cpp
@@ -203,8 +203,12 @@ void MemWorker::run() {
 	void *json_node = json_node_create(std::string(MODULE_NAME),
       		action_name.c_str(), rvs::loginfo);
 	if (json_node){
-		rvs::lp::AddString(json_node, "gpu_id",
-          		std::to_string(gpu_id));
+		rvs::lp::AddString(json_node, "gpu_id", std::to_string(gpu_id));
+
+    uint16_t gpu_index = 0;
+    rvs::gpulist::gpu2gpuindex(gpu_id, &gpu_index);
+    rvs::lp::AddString(json_node, "gpu_index", std::to_string(gpu_index));
+
 		rvs::lp::AddString(json_node,"Total Memory", std::to_string(total));
 		rvs::lp::AddString(json_node,"Free Memory", std::to_string(free));
 		rvs::lp::LogRecordFlush(json_node, rvs::logresults);
@@ -289,8 +293,12 @@ void MemWorker::log_to_json(int log_level, KVPairs...  key_values ) {
     void *json_node = json_node_create(std::string(MODULE_NAME),
         action_name.c_str(), log_level);
     if (json_node) {
-      rvs::lp::AddString(json_node, "gpu_id",
-          std::to_string(gpu_id));
+      rvs::lp::AddString(json_node, "gpu_id", std::to_string(gpu_id));
+
+      uint16_t gpu_index = 0;
+      rvs::gpulist::gpu2gpuindex(gpu_id, &gpu_index);
+      rvs::lp::AddString(json_node, "gpu_index", std::to_string(gpu_index));
+
       for (int i =0; i< kvlist.size()-1; i +=2){
           rvs::lp::AddString(json_node, kvlist[i], kvlist[i+1]);
       }

diff --git a/peqt.so/src/action.cpp b/peqt.so/src/action.cpp
@@ -203,8 +203,11 @@ bool peqt_action::get_gpu_all_pcie_capabilities(struct pci_dev *dev,
         action_name.c_str(), rvs::logresults);
 
       if (json_pcaps_node != NULL) {
-        rvs::lp::AddString(json_pcaps_node, RVS_JSON_LOG_GPU_ID_KEY,
-                std::to_string(gpu_id));
+        rvs::lp::AddString(json_pcaps_node, RVS_JSON_LOG_GPU_ID_KEY, std::to_string(gpu_id));
+
+        uint16_t gpu_index = 0;
+        rvs::gpulist::gpu2gpuindex(gpu_id, &gpu_index);
+        rvs::lp::AddString(json_pcaps_node, RVS_JSON_LOG_GPU_IDX_KEY, std::to_string(gpu_index));
       }
     }
     for (it = property.begin(); it != property.end(); ++it) {

diff --git a/perf.so/src/perf_worker.cpp b/perf.so/src/perf_worker.cpp
@@ -1,6 +1,6 @@
 /********************************************************************************
  *
- * Copyright (c) 2018-2022 Advanced Micro Devices, Inc. All rights reserved.
+ * Copyright (c) 2018-2024 Advanced Micro Devices, Inc. All rights reserved.
  *
  * MIT LICENSE:
  * Permission is hereby granted, free of charge, to any person obtaining a copy of
@@ -277,8 +277,12 @@ void PERFWorker::log_to_json(const std::string &key, const std::string &value,
         void *json_node = rvs::lp::LogRecordCreate(MODULE_NAME,
                             action_name.c_str(), log_level, sec, usec);
         if (json_node) {
-            rvs::lp::AddString(json_node, PERF_JSON_LOG_GPU_ID_KEY,
-                            std::to_string(gpu_id));
+            rvs::lp::AddString(json_node, PERF_JSON_LOG_GPU_ID_KEY, std::to_string(gpu_id));
+
+            uint16_t gpu_index = 0;
+            rvs::gpulist::gpu2gpuindex(gpu_id, &gpu_index);
+            rvs::lp::AddString(json_node, "gpu_index", std::to_string(gpu_index));
+
             rvs::lp::AddString(json_node, key, value);
             rvs::lp::LogRecordFlush(json_node);
         }

diff --git a/pesm.so/src/worker.cpp b/pesm.so/src/worker.cpp
@@ -262,6 +262,11 @@ void Worker::run() {
       r = json_node_create("PESM",
         stop_action_name.c_str(), rvs::logresults);
       rvs::lp::AddString(r, "gpu_id", std::to_string(i.first));
+
+      uint16_t gpu_index = 0;
+      rvs::gpulist::gpu2gpuindex(i.first, &gpu_index);
+      rvs::lp::AddString(r, "gpu_index", std::to_string(gpu_index));
+
       rvs::lp::AddString(r, "pcie speed change",i.second);
       rvs::lp::LogRecordFlush(r);
       r = nullptr;
@@ -278,6 +283,11 @@ void Worker::run() {
       r = json_node_create("PESM",
         stop_action_name.c_str(), rvs::logresults);
       rvs::lp::AddString(r, "gpu_id", std::to_string(i.first));
+
+      uint16_t gpu_index = 0;
+      rvs::gpulist::gpu2gpuindex(i.first, &gpu_index);
+      rvs::lp::AddString(r, "gpu_index", std::to_string(gpu_index));
+
       rvs::lp::AddString(r, "power_change", i.second);
       rvs::lp::AddString(r, "msg", "stopped");
       rvs::lp::LogRecordFlush(r);

diff --git a/tst.so/src/tst_worker.cpp b/tst.so/src/tst_worker.cpp
@@ -54,6 +54,7 @@
 #define TST_PASS_KEY                            "pass"
 
 #define TST_JSON_LOG_GPU_ID_KEY                 "gpu_id"
+#define TST_JSON_LOG_GPU_IDX_KEY                "gpu_index"
 #define TST_MEM_ALLOC_ERROR                     1
 #define TST_BLAS_ERROR                          2
 #define TST_BLAS_MEMCPY_ERROR                   3
@@ -95,6 +96,8 @@ void TSTWorker::log_to_json(const std::string &key, const std::string &value,
         if (json_node) {
             rvs::lp::AddString(json_node, TST_JSON_LOG_GPU_ID_KEY,
                             std::to_string(gpu_id));
+            rvs::lp::AddString(json_node, TST_JSON_LOG_GPU_IDX_KEY,
+                            std::to_string(smi_device_index));
             rvs::lp::AddString(json_node, key, value);
             rvs::lp::LogRecordFlush(json_node);
         }